Novel Three-Finger Neurotoxins from Naja melanoleuca Cobra Venom Interact with GABAA and Nicotinic Acetylcholine Receptors

Cobra venoms contain three-finger toxins (TFT) including α-neurotoxins efficiently binding nicotinic acetylcholine receptors (nAChRs). As shown recently, several TFTs block GABAA receptors (GABAARs) with different efficacy, an important role of the TFTs central loop in binding to these receptors being demonstrated. We supposed that the positive charge (Arg36) in this loop of α-cobratoxin may explain its high affinity to GABAAR and here studied α-neurotoxins from African cobra N. melanoleuca venom for their ability to interact with GABAARs and nAChRs. Three α-neurotoxins, close homologues of the known N. melanoleuca long neurotoxins 1 and 2, were isolated and sequenced. Their analysis on Torpedo californica and α7 nAChRs, as well as on acetylcholine binding proteins and on several subtypes of GABAARs, showed that all toxins interacted with the GABAAR much weaker than with the nAChR: one neurotoxin was almost as active as α-cobratoxin, while others manifested lower activity. The earlier hypothesis about the essential role of Arg36 as the determinant of high affinity to GABAAR was not confirmed, but the results obtained suggest that the toxin loop III may contribute to the efficient interaction of some long-chain neurotoxins with GABAAR. One of isolated toxins manifested different affinity to two binding sites on Torpedo nAChR.


Introduction
Ligand-gated ion channels are classified into several families of membrane-bound receptors. The Cys-loop family of pentameric receptors is represented widely in the muscle and nervous systems as well as in immune and other cells and plays prominent roles. The Cys-loop family of vertebrates includes nicotinic acetylcholine receptor (nAChR), as well as serotonin type 3, γ-aminobutyric acid (GABA A R), and glycine receptors. The receptors of this family are characterized by a conserved sequence of 13 amino acid residues confined by two cysteines forming a disulfide bond (Cys loop) in the N-terminal extracellular domain of each subunit [1]. Among these receptors, excitatory nAChRs are the targets of numerous natural compounds including toxins from snake venom [2,3]. Snake venom neurotoxins of molecular basis for the high selectivity of α-conotoxin LvIA for α3β2 nAChR [20] or the difference in affinity of two other α-conotoxins' analogues to human and rat α7 nAChR [21] were explained. At present, AChBPs from the mollusks Aplysia californica and Lymnaea stagnalis [22] are widely used, the former is closer pharmacologically to homooligimeric nAChRs while the latter is closer to heterooligomeric receptors.
α-Bgt, being a very efficient blocker of nAChRs, showed fairly weak activity on GABA A R inhibiting α1β3γ2 receptors by only 19% at 10 µM [15]. In contrast, α-Ctx inhibited GABA A R quite effectively manifesting half-maximal inhibitory concentration (IC 50 ) of 236 nM at α1β3γ receptor [15]. Some other snake neurotoxins inhibited GABA A R as well, although not so effectively as α-Ctx. It was suggested that Arg36, present in α-Ctx and being valine in this position of α-Bgt, might be responsible for efficient interaction of α-Ctx with GABA A R [6]. In order to test this hypothesis, α-Ctx analogues with or without Arg36 need to be used. Although the most straightforward way would be testing the α-Ctx analogue(s) obtained through site-directed mutagenesis at position occupied by Arg36, there are some problems that should be solved on this way. First one is that α-Ctx contains five disulfide bridges the correct formation of which in mutant should be proved. This may require either determination of spatial structure (e.g., by X-ray or NMR) or combined used of selective disulfide modification and peptide mass fingerprinting. Moreover, the cleavage of some chimeric proteins often used to obtain the protein of interest or direct expression of the protein may add extra amino acid residues at N-terminus, which may influence the biological activity. We believed that use of natural α-Ctx analogues containing either Arg36 or other residue at this position may be the easiest way to test our hypothesis. If natural α-Ctx analogue with Arg36 is not as active as α-Ctx itself, then this arginine residue is not the only determinant of high α-Ctx affinity to GABA A R. In this respect, our attention was attracted to the cobra N. melanoleuca venom in which the presence of two highly homologous long chain neurotoxins 1 and 2 was shown [23,24]: neurotoxin 1 contained valine and neurotoxin 2 contained arginine residue at the position corresponding to Arg36 in α-Ctx. They seemed a good pair to clarify the role the respective residue.
In the present work from N. melanoleuca venom we isolated three α-neurotoxins (two with arginine and one with valine at the position corresponding to Arg36 in α-Ctx) and one muscarinic toxin-like protein. Only one neurotoxin (designated here as TX-NM4) was identical to the earlier described N. melanoleuca neurotoxin 2, another (TX-NM3-1) was its analogue with 5 substitutions, while instead of neurotoxin 1 we isolated its analogue with 3 substitutions. Analyzing binding of the isolated toxins to GABA A Rs and inhibition of the ion currents elicited by GABA, we did not find any evidence for the role of Arg36. However, molecular modeling showed that α-Ctx loop III had contact with α1 subunit of GABA A R and this contact might contribute to the efficient binding of the toxin to the receptor. Thus, our work expands the range of TFTs capable of inhibiting the GABA A R subtypes. In addition, we found that one N. melanoleuca toxin (Tx-NM2) showed unequal affinities for the two ligand-binding sites in the Torpedo californica nAChR and thus widened the range of naturally occurring TFTs that differently interact with the two binding sites in muscle-type nAChRs. Keeping in mind the future modeling and structural studies of the interaction of N. melanoleuca toxins with nAChRs and GABA A Rs, we investigated the interaction of these toxins with AChBPs from A. californica and L. stagnalis as plausible models. Tx-NM2 was found to be the most potent in binding with both AChPBs.

Isolation and Characterization of Naja melanoleuca Toxins
To isolate toxins, a three-step chromatographic procedure was used. Gel filtration on Sephadex G50 was the first step ( Figure 1a). The analysis of obtained fractions by mass spectrometry showed that fraction 6 contained proteins with molecular masses in the range of 6-8 kDa. In cobra venom, these masses are characteristic of three-finger toxins. As mentioned in introduction, we were interested in long neurotoxins 1 and 2 possessing the molecular masses of 8040.76 and 7756.59 Da, respectively. Therefore, the fraction 6 containing three-finger toxins was further separated by ion exchange chromatography on HEMA BIO 1000CM column ( Figure 1b) and fractions obtained were analyzed by mass spectrometry. Fractions 2, 3, and 4 containing toxins with molecular masses of about 8 kDa were further purified by reversed phase HPLC ( Figure S1). As a result, one toxin was obtained from fraction 2 (Tx-NM2), two toxins from fraction 3 (Tx-NM3-1, Tx-NM3-2) and one toxin from fraction 4 (Tx-NM4). Finally, four toxins with molecular masses of 8030.68 (Tx-NM2), 7787.577 (Tx-NM3-1), 7441.497 (Tx-NM3-2), and 7756.581 (Tx-NM4) Da were purified (Table 1, Figure S2).

Isolation and Characterization of Naja melanoleuca Toxins
To isolate toxins, a three-step chromatographic procedure was used. Gel filtration on Sephadex G50 was the first step ( Figure 1a). The analysis of obtained fractions by mass spectrometry showed that fraction 6 contained proteins with molecular masses in the range of 6-8 kDa. In cobra venom, these masses are characteristic of three-finger toxins. As mentioned in introduction, we were interested in long neurotoxins 1 and 2 possessing the molecular masses of 8040.76 and 7756.59 Da, respectively. Therefore, the fraction 6 containing three-finger toxins was further separated by ion exchange chromatography on HEMA BIO 1000CM column ( Figure 1b) and fractions obtained were analyzed by mass spectrometry. Fractions 2, 3, and 4 containing toxins with molecular masses of about 8 kDa were further purified by reversed phase HPLC ( Figure S1). As a result, one toxin was obtained from fraction 2 (Tx-NM2), two toxins from fraction 3 (Tx-NM3-1, Tx-NM3-2) and one toxin from fraction 4 (Tx-NM4). Finally, four toxins with molecular masses of 8030.68 (Tx-NM2), 7787.577 (Tx-NM3-1), 7441.497 (Tx-NM3-2), and 7756.581 (Tx-NM4) Da were purified (Table 1, Figure S2).  The amino acid sequences of isolated toxins were determined by tandem mass spectrometry. For this purpose, the toxins were reduced, carbamidomethylated and digested with trypsin or chymotrypsin. The peptides obtained were analyzed by LC-MS with simultaneous de novo mass spectrometry sequencing. As a result, the complete amino acid sequences of Tx-NM2, Tx-NM3-1, and Tx-NM4 as well as partial amino acid sequence of Tx-NM3-2 were determined ( Figure 2).  The amino acid sequences of isolated toxins were determined by tandem mass spectrometry. For this purpose, the toxins were reduced, carbamidomethylated and digested with trypsin or chymotrypsin. The peptides obtained were analyzed by LC-MS with simultaneous de novo mass spectrometry sequencing. As a result, the complete amino acid sequences of Tx-NM2, Tx-NM3-1, and Tx-NM4 as well as partial amino acid sequence of Tx-NM3-2 were determined ( Figure 2).  Figure 2. Alignment of amino acid sequences. Long 1 is long type α-neurotoxin 1 from N. melanoleuca P01383 (3L21_NAJME), Long 2 is long type α-neurotoxin 2 from N. melanoleuca P01388 (3L22_NAJME), toxin P82462 (3SUC1_NAJKA) is muscarinic toxin like protein from N. kaouthia venom. All toxins have three loops stabilized by five disulfide bonds. The residues different in αneurotoxin 1 and Tx-NM2 as well as α-neurotoxin 2 and Tx-NM3-1 are underlined. Arginine and valine residues corresponding to position 36 in α-Ctx sequence are shown in lilac and green, respectively.
It was found that the amino acid sequence of Tx-NM4 exactly corresponds to that of long type α-neurotoxin 2 from N. melanoleuca (Uniprot accession number P01388). The sequence of Tx-NM2 differs from that of N. melanoleuca long type α-neurotoxin 1 (P01383) in three positions: at position 50, Lys residue is changed to Thr, Gln56 to Glu, and Met72 is oxidized. Tx-NM3-1 is very similar to long type α-neurotoxin 2 (P01388), but its sequences differ in 5 positions ( Figure 2). We were not able to determine the amino acid sequence of Tx-NM3-2 at the short C-terminal fragment, however this toxin is very similar to muscarinic toxin-like protein 1 (P82462) from N. kaouthia cobra venom ( Figure 2).

Interaction with GABAAR
The biological activity of isolated toxins was studied on rat GABAARs obtained by heterologous expression of different combination of α, β and γ subunits. The most common combinations of GABAAR subunits were used. They included α1β2γ2, α1β3γ2, and α3β2γ2 receptors. As concerns the inclusion of γ2 subunit, it has been shown earlier that increase in the ratio of mRNA/cDNA relative to α1 and β2 subunits (up to 10:1:1) resulted in more homogenous α1β2γ2 receptors in oocytes [25]. In our work, we used exactly this high γ2 subunit ratio relative α1 and β2 (10:1:1). Moreover, the presence of γ subunit in the receptors was proved in electrophysiology experiments as shown below.


Competition of N. melanoleuca toxins with fluorescently labeled α-Ctx In this work, we used the most prevalent types of GABAARs. We have shown previously that α-Ctx labelled with Alexa Fluor 546 fluorophore efficiently stained Neuro 2a cells heterologously expressing α1β3γ2 GABAAR and this staining was inhibited by native α-Ctx and some other toxins [15]. α-Ctx was shown to be the most active on this receptor type [15], and we used it first to test the activity of N. melanoleuca toxins ( Figure 3). The activity of N. melanoleuca toxins on the α1β3γ2 receptor was tested with the mentioned fluorescence assay. It was found that at 10 µM Tx-NM2 and Tx-NM3-1 inhibited Alexa Fluor 546 α-Ctx binding to GABAAR as efficiently as native α-Ctx, while Tx-NM4 was less active (Figure 3). Figure 2. Alignment of amino acid sequences. Long 1 is long type α-neurotoxin 1 from N. melanoleuca P01383 (3L21_NAJME), Long 2 is long type α-neurotoxin 2 from N. melanoleuca P01388 (3L22_NAJME), toxin P82462 (3SUC1_NAJKA) is muscarinic toxin like protein from N. kaouthia venom. All toxins have three loops stabilized by five disulfide bonds. The residues different in α-neurotoxin 1 and Tx-NM2 as well as α-neurotoxin 2 and Tx-NM3-1 are underlined. Arginine and valine residues corresponding to position 36 in α-Ctx sequence are shown in lilac and green, respectively.
It was found that the amino acid sequence of Tx-NM4 exactly corresponds to that of long type α-neurotoxin 2 from N. melanoleuca (Uniprot accession number P01388). The sequence of Tx-NM2 differs from that of N. melanoleuca long type α-neurotoxin 1 (P01383) in three positions: at position 50, Lys residue is changed to Thr, Gln56 to Glu, and Met72 is oxidized. Tx-NM3-1 is very similar to long type α-neurotoxin 2 (P01388), but its sequences differ in 5 positions ( Figure 2). We were not able to determine the amino acid sequence of Tx-NM3-2 at the short C-terminal fragment, however this toxin is very similar to muscarinic toxin-like protein 1 (P82462) from N. kaouthia cobra venom ( Figure 2). The biological activity of isolated toxins was studied on rat GABA A Rs obtained by heterologous expression of different combination of α, β and γ subunits. The most common combinations of GABA A R subunits were used. They included α1β2γ2, α1β3γ2, and α3β2γ2 receptors. As concerns the inclusion of γ2 subunit, it has been shown earlier that increase in the ratio of mRNA/cDNA relative to α1 and β2 subunits (up to 10:1:1) resulted in more homogenous α1β2γ2 receptors in oocytes [25]. In our work, we used exactly this high γ2 subunit ratio relative α1 and β2 (10:1:1). Moreover, the presence of γ subunit in the receptors was proved in electrophysiology experiments as shown below.
• Competition of N. melanoleuca toxins with fluorescently labeled α-Ctx In this work, we used the most prevalent types of GABA A Rs. We have shown previously that α-Ctx labelled with Alexa Fluor 546 fluorophore efficiently stained Neuro 2a cells heterologously expressing α1β3γ2 GABA A R and this staining was inhibited by native α-Ctx and some other toxins [15]. α-Ctx was shown to be the most active on this receptor type [15], and we used it first to test the activity of N. melanoleuca toxins ( Figure 3). The activity of N. melanoleuca toxins on the α1β3γ2 receptor was tested with the mentioned fluorescence assay. It was found that at 10 µM Tx-NM2 and Tx-NM3-1 inhibited Alexa Fluor 546 α-Ctx binding to GABA A R as efficiently as native α-Ctx, while Tx-NM4 was less active (Figure 3).

Electrophysiology measurements
Functional activities of the isolated toxins were studied by two-electrode voltage clamp on mouse GABAARs heterlogously expressed in Xenopus laevis oocytes (Figures 4  and 5). Three combinations of receptor subunits were used in this study: α1β2γ2, α3β2γ2, and α1β3γ2. The incorporation of γ subunit was confirmed in separate experiments which are shown for the α1β3γ2 receptor ( Figure 4c). GABA-induced currents at the α1β3γ2 receptor were not inhibited by Zn 2+ ions, while inhibition was observed at α1β3 receptors (not shown). The incorporation of γ subunit in α1β3γ2 receptors was further confirmed by their sensitivity to diazepam ( Figure 4c).
First, we checked the activity of toxins on the common receptor types α1β2γ2 and α3β2γ2, α1β2γ2 being the most common isoform ( Figure 4). Finally, the quantitative characteristics for binding (IC50) were determined at α1β3γ2 type ( Figure 5). On α1β2γ2 and α3β2γ2 receptors at concentration of 10 µM, Tx-NM3-1 almost completely inhibited GABA-evoked currents ( Figure 4b). Two other toxins Tx-NM4 and Tx-NM2 were less effective and Tx-NM3-2 showed extremely low activity ( Figure 4b). Interestingly, the efficiency of interactions of Tx-NM3-1, Tx-NM3-2, and Tx-NM4 with two GABAAR subtypes (α1β2γ2 and α3β2γ2) was different and the largest difference was observed for Tx-NM4 ( Figure 4b).

• Electrophysiology measurements
Functional activities of the isolated toxins were studied by two-electrode voltage clamp on mouse GABA A Rs heterlogously expressed in Xenopus laevis oocytes (Figures 4 and 5). Three combinations of receptor subunits were used in this study: α1β2γ2, α3β2γ2, and α1β3γ2. The incorporation of γ subunit was confirmed in separate experiments which are shown for the α1β3γ2 receptor ( Figure 4c). GABA-induced currents at the α1β3γ2 receptor were not inhibited by Zn 2+ ions, while inhibition was observed at α1β3 receptors (not shown). The incorporation of γ subunit in α1β3γ2 receptors was further confirmed by their sensitivity to diazepam ( Figure 4c).
First, we checked the activity of toxins on the common receptor types α1β2γ2 and α3β2γ2, α1β2γ2 being the most common isoform ( Figure 4). Finally, the quantitative characteristics for binding (IC 50 ) were determined at α1β3γ2 type ( Figure 5). On α1β2γ2 and α3β2γ2 receptors at concentration of 10 µM, Tx-NM3-1 almost completely inhibited GABAevoked currents ( Figure 4b). Two other toxins Tx-NM4 and Tx-NM2 were less effective and Tx-NM3-2 showed extremely low activity ( Figure 4b). Interestingly, the efficiency of interactions of Tx-NM3-1, Tx-NM3-2, and Tx-NM4 with two GABA A R subtypes (α1β2γ2 and α3β2γ2) was different and the largest difference was observed for Tx-NM4 ( Figure 4b).  More detailed studies of activity were performed at α1β3γ2 GABAAR. The dependences of GABA-elicited currents on toxin concentrations were investigated and IC50 values were determined. The inhibition curves for the most active toxins Tx-NM2 and Tx-NM3-1 are shown in Figure 5. The data for all four toxins studied are given in the Table 2. On this receptor subtype, Tx-NM3-1 was the most active, followed by Tx-NM2. Tx-NM4 was one order of to GABA without toxins is taken as 100%. Data are shown for α1β2γ2 and α3β2γ2 GABA A R and presented as Mean ± SEM (n = 3). For toxins Tx-NM3-1, Tx-NM3-2, and Tx-NM4, the difference in interaction with two receptor subtypes is statistically significant. p < 0.05 according to Student's t test.
(c) Example current traces at α1β2γ2 GABA A R illustrating gamma subunit incorporation. GABA potency was not affected by 50 µM of Zn 2+ but was enhanced by 1 µM of Diazepam (Dz).
More detailed studies of activity were performed at α1β3γ2 GABA A R. The dependences of GABA-elicited currents on toxin concentrations were investigated and IC 50 values were determined. The inhibition curves for the most active toxins Tx-NM2 and Tx-NM3-1 are shown in Figure 5.  More detailed studies of activity were performed at α1β3γ2 GABAAR. The dependences of GABA-elicited currents on toxin concentrations were investigated and IC50 values were determined. The inhibition curves for the most active toxins Tx-NM2 and Tx-NM3-1 are shown in Figure 5. The data for all four toxins studied are given in the Table 2. On this receptor subtype, Tx-NM3-1 was the most active, followed by Tx-NM2. Tx-NM4 was one order of  The data for all four toxins studied are given in the Table 2. On this receptor subtype, Tx-NM3-1 was the most active, followed by Tx-NM2. Tx-NM4 was one order of magnitude less active than the last two toxins, and Tx-NM3-2 manifested only very weak activity.
Summing up all the data concerning the action of the N. melanoleuca TFTs on GABA A Rs, we can conclude that Tx-NM3-1 was the most active, Tx-NM2 was a little bit less active, Tx-NM4 manifested very weak activity, and Tx-NM3-2 was practically inactive.

Interaction with nAChRs
The binding of isolated toxins to nAChRs was studied by competitive radioligand method using radioactive α-Bgt ( 125 I-αBgt) as a ligand. Interaction with nAChRs of muscle type from Torpedo electric organ and of human α7 type heterologously expressed in GH 4 C 1 cells was investigated. It was found that Tx-NM2, Tx-NM3-1, and Tx-NM4 efficiently inhibited binding to both receptor subtypes ( Figure 6). IC 50 values for studied toxins were in the nanomolar range (Table 3). magnitude less active than the last two toxins, and Tx-NM3-2 manifested only very weak activity. Summing up all the data concerning the action of the N. melanoleuca TFTs on GABAARs, we can conclude that Tx-NM3-1 was the most active, Tx-NM2 was a little bit less active, Tx-NM4 manifested very weak activity, and Tx-NM3-2 was practically inactive.

Interaction with nAChRs
The binding of isolated toxins to nAChRs was studied by competitive radioligand method using radioactive α-Bgt ( 125 I-αBgt) as a ligand. Interaction with nAChRs of muscle type from Torpedo electric organ and of human α7 type heterologously expressed in GH4C1 cells was investigated. It was found that Tx-NM2, Tx-NM3-1, and Tx-NM4 efficiently inhibited binding to both receptor subtypes ( Figure 6). IC50 values for studied toxins were in the nanomolar range (Table 3).  Interestingly, at Torpedo nAChR, the experimental data for Tx-NM2 differed greatly from those for Tx-NM3-1 and Tx-NM4 and fit two-site binding model with fairly smooth inhibition curve (Figure 6a). The difference in affinity of Tx-NM2 to these binding sites was about one order of magnitude (1 nM versus 8.66 nM) ( Table 3). The affinities of Tx-NM3-1 and Tx-NM4 to Torpedo nAChR were almost similar, Tx-NM3-1 being a slightly  Interestingly, at Torpedo nAChR, the experimental data for Tx-NM2 differed greatly from those for Tx-NM3-1 and Tx-NM4 and fit two-site binding model with fairly smooth inhibition curve (Figure 6a). The difference in affinity of Tx-NM2 to these binding sites was about one order of magnitude (1 nM versus 8.66 nM) ( Table 3). The affinities of Tx-NM3-1 and Tx-NM4 to Torpedo nAChR were almost similar, Tx-NM3-1 being a slightly more active. At the α7 nAChR, the most active was Tx-NM3-1 with IC 50 value of 4.84 nM, while Tx-Nm2 and Tx-NM4 were less active manifesting IC 50 values of 13.02 and 26.89 nM, respectively (Figure 6b). At a concentration of 1 µM, toxin Tx-NM3-2 inhibited 125 I-αBgt binding by only about 40% and was not analyzed further.

Interaction with Acetylcholine Binding Proteins from Lymnaea stagnalis and Aplysia californica
Competitive radioligand assay with 125 I-αBgt was applied also for testing the interaction of N. melanoleuca toxins with acetylcholine binding proteins (AChBPs) from L. stagnalis and A. californica. The results obtained showed that Tx-NM2 was the most active ( Figure 7). The affinity of this toxin to both proteins exceeded that of α-Bgt and α-Ctx (Table 4). However, the affinity of Tx-NM3-1 and Tx-NM4 to both proteins was the lowest among the toxins studied ( Figure 7, Table 4).  (Figure 6b). At a concentration of 1 µM, toxin Tx-NM3-2 inhibited 125 I-αBgt binding by only about 40% and was not analyzed further.

Interaction with Acetylcholine Binding Proteins from Lymnaea stagnalis and Aplysia californica
Competitive radioligand assay with 125 I-αBgt was applied also for testing the interaction of N. melanoleuca toxins with acetylcholine binding proteins (AChBPs) from L. stagnalis and A. californica. The results obtained showed that Tx-NM2 was the most active ( Figure 7). The affinity of this toxin to both proteins exceeded that of α-Bgt and α-Ctx (Table 4). However, the affinity of Tx-NM3-1 and Tx-NM4 to both proteins was the lowest among the toxins studied ( Figure 7, Table 4).

Molecular Modelling
The efficiency of Tx-NM3-1 and Tx-NM4 interactions with α1β3γ2 GABAAR differs more than an order of magnitude (Table 2). However, the amino acid sequences of these toxins differ only in five positions-three located at the tips of loops II and two at the tip of loop III. This may suggest the involvement of the loop III in interaction with GABAAR, in addition to loop II the importance of which for binding was shown earlier [15]. Interestingly, the amino acid sequences of Tx-NM3-1 and α-Ctx are practically identical in this loop III region: positions 49-58 in Figure 8. To find structural elements which may explain

Molecular Modelling
The efficiency of Tx-NM3-1 and Tx-NM4 interactions with α1β3γ2 GABA A R differs more than an order of magnitude (Table 2). However, the amino acid sequences of these toxins differ only in five positions-three located at the tips of loops II and two at the tip of loop III. This may suggest the involvement of the loop III in interaction with GABA A R, in addition to loop II the importance of which for binding was shown earlier [15]. Interestingly, the amino acid sequences of Tx-NM3-1 and α-Ctx are practically identical in this loop III region: positions 49-58 in Figure 8. To find structural elements which may explain the differences in the interaction of N. melanoleuca toxins with GABA A R, we performed the molecular modeling of toxin spatial structures and structure of the complex between GABA A R and toxin. the differences in the interaction of N. melanoleuca toxins with GABAAR, we performed the molecular modeling of toxin spatial structures and structure of the complex between GABAAR and toxin.   The models of N. melanoleuca toxin structures were constructed by homology modeling in Swissmodel service using α-Ctx structure (PDB 1YI5) as a template. The loop III structures for N. melanoleuca toxins are quite similar (Figure 9a). the differences in the interaction of N. melanoleuca toxins with GABAAR, we performed the molecular modeling of toxin spatial structures and structure of the complex between GABAAR and toxin.   Resulting homology models were submitted to protein docking Tox Dock instrument on the Rosetta server [26,27] to generate plausible starting structures of the toxins-GABA A R complexes. Flexibility of toxin molecules and receptors were taken into account and model structures were subsequently subjected to short molecular dynamics using GROMACS 5.0 package [28]. The resulting structures were visualized and inspected in UCSF Chimera [29] ( Figure 10). All models showed similar positions with the TFT's loop II buried under the C-loop of beta3-subunit of GABA A R (Figure 10a-e for the side view and Figure 10f-j for the top view from the extracellular side). Modeled complexes were compared to published X-ray structures of α-Bgt in complex with α7 nAChR/AChBP chimera [30] (Figure 10l) and of α-Ctx in complex with AChBP [31] (Figure 10m) as well as cryo-EM structures of α-Bgt in complex with muscle nAChR [32] (Figure 10k). Structures of principal subunits (α1 and α7 nAChRs, AChBP(+) and β3 GABA A R) of all mentioned above complexes were aligned in UCSF Chimera and TFT positions relative to the receptor were inspected. Interestingly, all putative complexes with GABA A R showed TFT position tilted towards the complementary subunit as compared to nAChR complexes (Figure 10n). In such tilted position the loop III of TFT molecule can form contacts with the complementary subunit in the GABA A R binding site.
of the α1 subunit. (c) Sequence alignment of N. melanoleuca neurotoxins with α-Ctx and α-Btx, based on alignments of spatial structures in UCSF Chimera "match-align" tool.
Resulting homology models were submitted to protein docking Tox Dock instrument on the Rosetta server [26,27] to generate plausible starting structures of the toxins-GABAAR complexes. Flexibility of toxin molecules and receptors were taken into account and model structures were subsequently subjected to short molecular dynamics using GROMACS 5.0 package [28]. The resulting structures were visualized and inspected in UCSF Chimera [29] (Figure 10). All models showed similar positions with the TFT's loop II buried under the C-loop of beta3-subunit of GABAAR (Figure 10a-e for the side view and Figure 10f-j for the top view from the extracellular side). Modeled complexes were compared to published X-ray structures of α-Bgt in complex with α7 nAChR/AChBP chimera [30] (Figure 10l) and of α-Ctx in complex with AChBP [31] (Figure 10m) as well as cryo-EM structures of α-Bgt in complex with muscle nAChR [32] (Figure 10k). Structures of principal subunits (α1 and α7 nAChRs, AChBP(+) and β3 GABAAR) of all mentioned above complexes were aligned in UCSF Chimera and TFT positions relative to the receptor were inspected. Interestingly, all putative complexes with GABAAR showed TFT position tilted towards the complementary subunit as compared to nAChR complexes (Figure 10n). In such tilted position the loop III of TFT molecule can form contacts with the complementary subunit in the GABAAR binding site.  The positions toxin loops II placed to the orthosteric GABA A R ligand binding site show a remarkable correlation with the functional inhibitory activity of the modeled toxins. As follows from our measurements, all tested TFTs can be ranked according to their inhibitory activity toward GABA A R in the following way: α-Ctx = Tx-NM3-1 > Tx-NM2 > Tx-NM4 > α-Bgt. On the other hand, distances between β-carbons of Thr 202 located at the tip of loop C of β3 subunit and Arg 65 located on the complementary side of the orthosteric ligand binding site of the receptor α1 subunit are as follows: for α-Ctx-17 Å, Tx-NM3-1-15 Å, Tx-NM2-14 Å, Tx-NM4-13 Å, and α-Bgt-12 Å. The distance between these atoms may be used as a simple metric of loop C closure upon ligand binding and may be a possible predictor of the inhibitory activity of the TFT. The stronger toxin is bound to receptor and the deeper inserted in the binding pocket, the longer is distance between the indicated amino acid residues. However, this may not be true for low molecular ligands. Thus, cryo-EM structure of GABA A R with non-peptidic antagonist bicuculline (PDB 6HUK) is characterized by the even shorter distance between β-carbons of the above mentioned Thr and Arg residues, that is only 10 Å.
According to molecular modeling of the α-Ctx-GABA A R complex performed earlier [15], loop III contacts GABA A R through Asp56 salt bridges with Arg65 and Arg171 of the α1 subunit and Thr53 forming hydrogen bond with Val177 backbone of the α1 subunit. Asp56 is present in the amino acid sequence of Tx-NM3-1 but is replaced by glutamic acid in α-Bgt and Tx-NM2 and by asparagine in Tx-NM4 (Figure 9b,c). Thr53 present in α-Ctx and Tx-NM3-1 is replaced by proline in α-Bgt, Tx-Nm2, and Tx-Nm4. These replacements might substantially weaken the interaction of toxins with GABA A R. However, further more extensive molecular modeling and structure-function relationship studies are needed to investigate details of TFT-GABA A R pharmacophore in details.

Discussion
Earlier it was shown that several snake venom toxins blocked GABA A R [13][14][15]. We have found that the snake venom toxins manifested a different efficiency of the interaction with the receptor, α-Ctx being the most active [15]. As mentioned in the introduction, we suggested that the Arg36 present in α-Ctx (Arg39 in the alignment given in Figure 2) might determine the highest activity of this toxin against GABA A R. To check if this is true, we decided to compare the activity of toxins which differ at this position and for this purpose isolated toxins from N. melanoleuca cobra venom. Two long type α-neurotoxins were isolated earlier from this venom: long neurotoxin 1 (P01383) [24] containing valine residue at the position 38 (Figure 2), and long neurotoxin 2 (P01388) [23] with Arg38 in the sequence (Figure 2). The only information about biological activity available for these toxins was LD 50 of 1.5 mg/kg for neurotoxin 1 by intraperitoneal injection [24]. We wanted to get these particular toxins, but in our work we isolated 4 toxins with molecular masses in the range of 7-8 kDa, characteristic for long type α-neurotoxins (Table 1). Their analysis by high resolution mass spectrometry showed that one of the isolated toxins, designated by us as Tx-NM4, was identical to long neurotoxin 2, while toxin Tx-NM3-1 was long neurotoxin 2 analogue with substitutions in five positions (Figure 2). Tx-NM2 represented an analogue of N. melanoleuca long neurotoxin 1 (P01383) and Tx-NM3-2 was homologous to muscarinic toxin-like protein 1 (MTLP-1, P82462) from N. kaouthia cobra venom. The differences between the published amino acid sequences and those determined in the present work may be explained by different geographical origin of snakes from which the venoms were obtained.
The biological activity of the isolated toxins was assessed against several molecular targets. In the studies on GABA A R, competition experiments with fluorescently labeled α-Ctx and electrophysiological assays were carried out using several combinations of receptor subunits. Both methods gave the consistent results. Toxin Tx-NM3-1 was the most active on all GABA A R subtypes studied and manifested activity similar to that of α-Ctx. However, at α1β3γ2 receptor subtype Tx-NM3-1 was less active than α-Ctx displaying IC 50 of 680 nM in contrast to 236 nM for α-Ctx [15]. Tx-NM2 was slightly less active than Tx-NM3-1 showing IC 50 of 1.25 µM at α1β3γ2 GABA A R. Tx-NM4 was a weaker antagonist than Tx-NM2 and Tx-NM3-1 at all subunit combinations, while Tx-NM3-2 was practically inactive. At the inhibition of α1β3γ2 GABA A R, the difference in affinity between the most active Tx-NM3-1 and the least active Tx-NM4 was more than one order of magnitude. Tx-NM4 at 10 µM more potently inhibited α1β2γ2 than α3β2γ2 GABA A R. Thus, the earlier suggestion about essential role of Arg36 (Arg39 in Figures 2 and 8) in α-Ctx binding to GABA A R was not confirmed, as both Tx-NM3-1 and Tx-NM4 contain arginine residue at this position. The analysis of toxin amino acid sequences revealed the identity of loop III sequence in α-Ctx and Tx-NM3-1. Molecular modeling of α-Ctx-GABA A R complex revealed the contacts between loop III of α-Ctx and α1 subunit of GABA A R. It was found that Asp56 of α-Ctx loop III forms salt bridges with Arg65 and Arg171 of the α1 subunit, and Thr53 forms a hydrogen bond with Val177 backbone of the α1 subunit (Figure 9b). The aspartic acid and threonine residues are present at these positions in Tx-NM3-1 as well. The interaction of these residues with α1 subunit of GABA A R may contribute to the stronger binding of α-Ctx and Tx-NM3-1 to the receptor.
Tx-NM3-2 is the muscarinic toxin-like protein and its amino acid sequence differs greatly from those of the long type neurotoxins to which other studied toxins belong. The muscarinic toxin-like protein was tested on GABA A R for the first time and it showed some activity, albeit very weak, against α3β2γ2 GABA A R (Figure 4). This is the first indication of the activity of muscarinic toxin-like protein against GABA A R.
The activity of all isolated toxins was studied in competition experiments for binding to Torpedo and α7 nAChRs as well as to two AChBPs. The neurotoxins tested showed a relatively high affinity to both receptor subtypes. It was found that at Torpedo nAChR, Tx-NM2 distinguished two binding sites, the affinity differing by about an order of magnitude (Table 3). Earlier, it was shown that the two binding sites in Torpedo and muscle types nAChRs were distinguished by αδ-bungarotoxins from the krait B. candidus venom [12], the difference being 17-fold for αδ-bungarotoxin-1. Comparison of the amino acid sequences of N. melanoleuca toxins with those of bungarotoxins and α-Ctx showed that the sequence of Tx-NM2 has a higher identity to that of αδ-bungarotoxin-1 than to α-Btx and α-Ctx (69% versus 61 and 55%, respectively, Figure 8). At the same time Tx-NM3-1 has 81% identical residues with α-Ctx and only 50% with αδ-bungarotoxin-1. Moreover, the sequence characteristics that were supposed to be responsible for difference in activity between α-Btx and αδ-bungarotoxin-1, i.e., the shortened loop I, the change of Phe residue in position 32 to Trp as well as Arg25 to Thr, are present in Tx-NM2. In addition, only Tx-NM2 and αδ-bungarotoxin-1 contain Tyr residue in position 4 and a positively charged residue in position 5. This similarity in structural features and activity between Tx-NM2 and αδbungarotoxin-1 supports the earlier ideas [12] on the capacity of long chain neurotoxin to distinguish two binding sites in the muscle-type nAChRs. It was found that Tx-NM3-1 was the most active against α7 nAChR (IC 50 4.84 nM), followed by Tx-NM2 and Tx-NM4 with IC 50 of 13.02 and 26.89 nM, respectively. Although the affinities of these toxins for nAChR are much higher than for GABA A R, their rank coincides with that observed for GABA A R suggesting that both receptors might have similar structural elements involved in toxin binding to them.
AChBPs are considered as models of the extracellular domains of Cys-loop receptors, in particular of nAChRs. Interestingly, AChBPs has only 20-24% sequence identity with nAChRs, however their pharmacological properties are similar to those of nAChRs. Studies of the binding of N. melanoleuca toxins to AChBPs from L. stagnalis and A. californica showed that toxin Tx-NM2 possessed the highest affinity to both proteins, while toxins Tx-NM3-1 and Tx-NM4 were several orders of magnitude less active. On A. californica AChBP, Tx-NM3-1 was the least active, which is very different from its activity against nAChR and GABA A R, where it was most active. This once again highlights the differences between AChBPs and the extracellular domains of nAChR and the GABA A R.
Considering the above data for nAChRs and AChBPs, one can conclude that for these targets, N. melanoleuca toxins manifest affinities similar to those of α-Btx and α-Ctx, and Tx-NM2 on AChBP from L. stagnalis showed the affinity higher than these two neurotoxins.
To obtain some information about possible molecular mechanisms determining the specificity of TFT interaction with GABA A R, we performed molecular modeling of the complexes formed by N. melanoleuca toxins, α-Bgt and α-Ctx with GABA A R. The modeling showed that the orientation of toxin molecules at GABA A R and nAChRs were different and toxin molecules in complexes with GABA A R tilted towards the complementary subunit. This tilt can lead to close contact of toxin loop III with the α1 subunit of the receptor. More detailed consideration revealed the amino acid residues which can form salt bridges (Asp56 in α-Ctx) and hydrogen bond (Thr53 in α-Ctx) with α1 subunit. These amino acid residues are present in the amino acid sequence of Tx-NM3-1, interacting with GABA A R with affinity similar to that of α-Ctx, but are replaced by other residues in less active toxins. Such substitutions may lead to a significant decrease in the efficiency of toxin interactions with GABA A R.

Conclusions
In our previous paper [15] it was found that among several snake toxins studied only α-Ctx manifested high activity against GABA A R and we put forward the hypothesis about essential role of Arg36 as the determinant of high affinity to GABA A R. To check this hypothesis, in this work we additionally studied several snake toxins for their ability to interact with GABA A R. For this purpose, four toxins were isolated from African cobra N. melanoleuca venom and their amino acid sequences were established by mass spectrometry. The amino acid sequence of one toxin was identical to that of previously known N. melanoleuca long neurotoxin 2. The second toxin sequenced differed from that of neurotoxin 2 in five positions. The third one was homologous to N. melanoleuca long neurotoxin 1; its sequence differed from that of neurotoxin 1 in three positions. One more toxin was homologous to muscarinic toxin-like protein from N. kaouthia venom and this is the first muscarinic toxin-like protein isolated from African cobra venom. The interaction of N. melanoleuca toxins with Torpedo and α7 nAChRs as well as with AChBPs and several subtypes of GABA A Rs was studied. One of isolated toxins, being the most active on AChBPs, manifested different affinity to two binding sites on Torpedo nAChR. Together with earlier data [12] this may indicate that there is a group of long type α-neurotoxins capable to bind with different affinity to two binding sites in the muscle type nAChR. All N. melanoleuca toxins interacted with the GABA A R much weaker than with the nAChR: one neurotoxin was almost as active as α-Ctx, while others manifested lower activity. The earlier hypothesis about essential role of Arg36 as the sole determinant of high toxin affinity to GABA A R was not confirmed, but the results of molecular modeling suggest that the loop III may contribute to the efficient interaction of some long-chain neurotoxins with GABA A R. Experimental proof of the modeling data will be the task of our future work.

Materials
All salts obtained from local suppliers were of analytical grade or higher. Venom of cobra N. melanoleuca was from Latoxan (Valence, France). Acetonitrile was purchased from Catrosa Reaktiv LLC (Moscow, Russia), and trifluoroacetic acid from Merck KGaA (Darmstadt, Germany). GH 4 C 1 cells transfected with hα7 nAChR cDNA were a gift of the Eli-Lilly Co. (London, UK). Muscle-type nAChR-enriched membranes from the electric organs of Torpedo californica were kindly provided by Prof. F. Hucho (Free University of Berlin, Germany). Acetylcholine binding proteins from L. stagnalis and A. californica were from Prof. A.B. Smit (Faculty of Earth and Life Sciences, Vrije Universiteit, Amsterdam).

Isolation of Neurotoxins
A 600 mg sample of dried N. melanoleuca venom was dissolved in 0.1 M ammonium acetate buffer, pH 6.2, and applied to a Sephadex G50s column (4.5 × 150 cm) equilibrated in the same buffer. The column was eluted at flow rate 32 mL/min. The fractions obtained were pooled as shown in Figure 1a. Fraction 6 was further separated on a HEMA BIO 1000CM column (4 × 250 mm) (Tessek, Prague, Czech Republic) in an ammonium acetate gradient from 5 to 500 mM (pH 7.5) in 100 min at flow rate 1.0 mL/min (Figure 1b). Fractions 2, 3 and 4 were freeze-dried and further purified by reversed phase chromatography on Jupiter C18 column (10 × 250 mm, Phenomenex, Torrance, CA, USA) in in a gradient of acetonitrile 20-35% in 60 min in the presence of 0.1% trifluoroacetic acid, at a flow rate of 2.0 mL/min ( Figure S1). After freeze-drying, the obtained proteins were used for further studies.
MS raw files were analyzed by PEAKS Studio 8.5 (Waterloo, ON, Canada) [35] and peak lists were searched against Serpentes Uniprot-Tremble FASTA (canonical and isoform) database version of May 2018 (144954 entries) with cysteine carbamidomethylation as a fixed modification and methionine oxidation and asparagine and glutamine deamidation as variable modifications. Enzyme specificity in the database search was set to trypsin with semi-specific digest mode. False discovery rate was set to 0.01 for peptide-spectrum matches and was determined by searching a reverse database. Peptide identification was performed with an allowed initial precursor mass deviation up to 10 p.p.m. and an allowed fragment mass deviation 0.05 Da.

Expression of GABAAR in Xenopus Oocytes
Xenopus laevis oocytes were prepared as described [36]. The work with oocytes was approved by the Shemyakin-Ovchinnikov Institute of Bioorganic Chemistry RAS with the approval number IACUC 251/2018 26.02.18. Plasmid DNAs encoding rat GABA A receptor subunits in pCI mammalian expression vector (Promega, Madison, WI, USA) were kindly provided by Dr. M. Ernst from the Medical University of Vienna. Next day after harvesting oocytes were injected by means of Nanoject II (Drummond Scientific, Broomall, PA, USA) with 2 ng mixture containing vector DNAs encoding receptor α1 or 3, β2 or 3, and γ2 subunits at 1:1:10 mass ratio. Injected oocytes were incubated at 18 • C for 2-3 days in ND96 solution (5 mM HEPES/NaOH pH 7.4, 96 mM NaCl, 2 mM KCl, 1.8 mM CaCl 2 , 2 mM MgCl 2 ) supplemented with gentamycin at 40 µg/mL.

Two-Electrode Voltage Clamp Electrophysiology Assay
Oocyte was placed in the flow through chamber combined with a nylon grid holding the bath of ND96 solution. Membrane potential of oocyte was clamped at −60mV by TURBO TEC-03X (npi electronic GmbH, Tamm, Germany). Electrodes were pulled from borosilicate capillary (Warner Instruments, Holliston, MA, USA) and filled with 3M KCl solution. Currents were recorded and digitized by WinWCP (University of Strathclyde, Glasgow, UK) software. After stable amplitude of 10 µM GABA-evoked currents was obtained oocyte was pre-incubated with toxin ND96-based solution for three minutes followed by co-application with 10 µM GABA. After steady baseline potential was achieved usually in a five-minute washout session oocyte was perfused with GABA again for current amplitude stability control purposes and pre-incubated with next sample. All solutions were applied manually with automatic pipette in a volume of 200 µL (Eppendorf, Hamburg, Germany). To confirm γ subunit incorporation, GABA response in the presence of 50 µM Zn 2+ and 1 µM diazepam was tested under the same conditions. Data for toxins are presented as the amplitude ratio of the currents elicited by 10 µM GABA in the presence of toxin to the average of the currents in control elicited by 10 µM GABA before co-application with toxin (100% control current amplitude) in the same oocyte. Data were collected from at least three oocytes from three different batches. Representative current traces, bar graphs and toxin-dependent dose response curves were plotted by Origin 8.1 (OriginLab, Northampton, MA, USA).

Fluorescent Ligand Competition Assay
Neuro2a cells sub-cultured 1:5 24 h before transfection were growing on clear 96well plate (Corning, Corning, NY, USA) in a complete DMEM (Paneco, Moscow, Russia). Lipofectamine (Invitrogen, Waltham, MA, USA) mediated transfection was performed with equal amounts (0.14 µg/mL) of pCI plasmid expression vectors encoding rat α1, β3, γ2 GABA A R subunits. Transfected Neuro2a cells were grown at 37 • C in 5% CO 2 incubator for 72 h, at the day of experiment medium was substituted for extracellular solution containing (in mM) 140 NaCl, 2 CaCl 2 , 2.8 KCl, 4 MgCl 2 , 20 HEPES, 10 glucose at pH 7.4. Cells were pre-incubated with 10 µM of toxins for 15 min at room temperature followed by 20 min of incubation with 50 nM Alexa Fluor 546 α-Ctx conjugate in the final volume of 100 µL. Afterwards cells were washed 3 times with two-fold excess of extracellular solution. To control the level of non-specific fluorescence experiment with 10 µM α-Ctx was run under the same conditions. By means of epifluorescent microscope IX71 (Olympus, Tokyo, Japan) equipped with CCD camera pictures of 3 fields chosen on each plate well in bright-field illumination were taken. Fluorescence was counted using CellX and ImageJ open-source software. Intensity of the fluorescence was normalized on integral intensity of the plate well incubated in presence of 50 nM Alexa Fluor 546 α-Ctx conjugate. Each experimental point is an average of integral intensity independently measured on 6 plate wells from three separate passages ± SEM.

Molecular Modeling
The molecular models of N. melanoleuca toxin structures were constructed using free bioinformatic tool Swissmodel (https://swissmodel.expasy.org/ (accessed on 12 February 2021)). α-Ctx structure (PDB 1YI5) was used as template. The molecular model of α-Ctx-GABA A R complex was taken from [15]. Briefly, the model of orthosteric binding site was constructed basing on extracellular domain from X-ray structure of β3 GABA A R subunit [39]. Swissmoldel service was used to build the extracellular domain of α1 GABA A R subunit. RMSD between modelled α1 extracellular domain and recently published cryo-EM structure was calculated by UCSF Chimera Match maker tool and did not exceed 1.2 Å. Homology models were submitted to protein docking "Tox Dock" instrument on the Rosetta server [26,27] to generate structures of the toxin-GABA A R complexes. Flexibility of toxin molecules and receptors were taken into account and model structures were subsequently subjected to short molecular dynamics using GROMACS 5.0 package. Briefly, putative structures of TFT-GABA A R complexes obtained via "Tox Dock" were energy minimized using steepest descent minimization algorithm to a maximum force <1000 kJ/mol/nm. Short-range electrostatic and Van der Waals cut-offs were set to 1 Å, periodic boundary conditions were constructed with rhombododecaedron simulation box having 1.2 Å from protein image to the closest boundary. After energy minimization two consequent constrained 100 ps molecular dynamics were performed to equilibrate systems in NVT (constant number of particles, volume, and temperature) and NPT (constant number of particles, pressure, and temperature) conditions. Equilibration was followed by 100 ps of unconstrained molecular dynamics [28]. The resulting structures were visualized and inspected in UCSF Chimera [29]. To compare TFT positions "Match" and "Match-align" functions of UCSF Chimera were used.

Informed Consent Statement: Not applicable.
Data Availability Statement: All data is contained within this article and supplementary materials.